Scholarlink

<p>Track-before-detect (TBD) filters improve target detection in low signal-to-noise ratio radar scenes, but remain sensitive to model mismatch between the assumed motion model and the actual target dynamics. We propose an adaptive regularization approach for the Bernoulli particle filter TBD that addresses this limitation by adjusting the kernel bandwidth applied after resampling. Two complementary indicators guide the bandwidth control: the bin occupancy count obtained from the spatial histogram of particles and the Bernoulli existence probability that reflects measurement consistency. When either indicator signals a mismatch, the bandwidth widens to maintain particle diversity, and it narrows again once the motion realigns with the model. Simulation results demonstrate that the proposed method maintains robust tracking performance under both matched and mismatched motion models while effectively preventing divergence during maneuvers.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Adaptive Regularization for Bernoulli Particle Filter Track-Before-Detect Under Motion Model Mismatch

  • 期刊论文

  • 发表日期：

    2026年7月20日

查看全文
  • Jehwa Lee,
  • Jae Hong Lee,
  • Du Yong Kim,
  • Chan Gook Park

摘要

Track-before-detect (TBD) filters improve target detection in low signal-to-noise ratio radar scenes, but remain sensitive to model mismatch between the assumed motion model and the actual target dynamics. We propose an adaptive regularization approach for the Bernoulli particle filter TBD that addresses this limitation by adjusting the kernel bandwidth applied after resampling. Two complementary indicators guide the bandwidth control: the bin occupancy count obtained from the spatial histogram of particles and the Bernoulli existence probability that reflects measurement consistency. When either indicator signals a mismatch, the bandwidth widens to maintain particle diversity, and it narrows again once the motion realigns with the model. Simulation results demonstrate that the proposed method maintains robust tracking performance under both matched and mismatched motion models while effectively preventing divergence during maneuvers.